About the Eager API support

[Daisuke Kato]

I've been using TensorFlow and Rust for a while now, and I'm very happy with this binding. Since this crate implements many APIs other than inference, I wanted to have Eager API as well. As far as I know, cppflow has partial support for the Eager API op, so I think I can make a basic implementation by referring to that and tensorflow's eager/c_api_test.

I think there are three major tasks required for this implementation.

* [ ] Creating bindings in `tensorflow-sys`.

* [ ] Implementing the Eager API

* [ ] Implementation of raw op (refer to cppflow)

At present, I believe that I can implement an API that can handle both Tensor and TensorHandle relatively transparently by leaving the Tensor struct as it is and implementing a common trait between Tensor and the newly implemented TensorHandle.

I myself often use TensorFlow for image-related tasks, and the Eager

API will be useful for basic I/O without relying on image crate.

If you have any suggestions or comments, I would appreciate your feedback.

Thank you.